Atlas

Large Language Models

The Transformer

The general-purpose block the whole field is built on.

You can jump straight in, but this star assumes Attention and Tokenization. Not recommended as a first stop.

Four layers of depth

Each layer ends with a quiz. Finish layer 4 and you own this concept.

  1. L1IntuitionHow attention, an MLP, residual connections, and normalization combine into 'the block' that every LLM is just a stack of.6m
    40 XP
  2. L2MechanicsPre-LN vs post-LN placement, encoder/decoder vs decoder-only architectures, and how the pieces assemble into a full model.10m
    70 XP
  3. L3CodeAssemble a full GPT-style transformer block and forward pass from scratch in PyTorch, and verify parameter counts against a real model.15m
    110 XP
  4. L4FoundationsThe full parameter and FLOP accounting for a dense transformer, and why arithmetic intensity makes autoregressive decoding fundamentally different from training.13m
    180 XP

Where this leads

33 stars in the atlas.